For Monday Finish chapter 19 Take-home exam due

Program 4 Any questions?

Beyond a Single Learner Ensembles of learners work better than individual learning algorithms Several possible ensemble approaches: –Ensembles created by using different learning methods and voting –Bagging –Boosting

Bagging Random selections of examples to learn the various members of the ensemble. Seems to work fairly well, but no real guarantees.

Boosting Most used ensemble method Based on the concept of a weighted training set. Works especially well with weak learners. Start with all weights at 1. Learn a hypothesis from the weights. Increase the weights of all misclassified examples and decrease the weights of all correctly classified examples. Learn a new hypothesis. Repeat

Approaches to Learning Maintaining a single current best hypothesis Least commitment (version space) learning

Different Ways of Incorporating Knowledge in Learning Explanation Based Learning (EBL) Theory Revision (or Theory Refinement) Knowledge Based Inductive Learning (in first-order logic - Inductive Logic Programming (ILP)

Explanation Based Learning Requires two inputs –Labeled examples (maybe very few) –Domain theory Goal –To produce operational rules that are consistent with both examples and theory –Classical EBL requires that the theory entail the resulting rules

Why Do EBL? Often utilitarian or speed-up learning Example: DOLPHIN –Uses EBL to improve planning –Both speed-up learning and improving plan quality

Theory Refinement Inputs the same as EBL –Theory –Examples Goal –Fix the theory so that it agrees with the examples Theory may be incomplete or wrong

Why Do Theory Refinement? Potentially more accurate than induction alone Able to learn from fewer examples May influence the structure of the theory to make it more comprehensible to experts

How Is Theory Refinement Done? Initial State: Initial Theory Goal State: Theory that fits training data. Operators: Atomic changes to the syntax of a theory: –Delete rule or antecedent, add rule or antecedent –Increase parameter, Decrease parameter –Delete node or link, add node or link Path cost: Number of changes made, or total cost of changes made.

Theory Refinement As Heuristic Search Finding the “closest” theory that is consistent with the data is generally intractable (NP­hard). Complete consistency with training data is not always desirable, particularly if the data is noisy. Therefore, most methods employ some form of greedy or hill­climibing search. Also, usually employ some form of over­fitting avoidance method to avoid learning an overly complex revision.

Theory Refinement As Bias Bias is to learn a theory which is syntactically similar to the initial theory. Distance can be measured in terms of the number of edit operations needed to revise the theory (edit distance). Assumes the syntax of the initial theory is “approximately correct.” A bias for minimal semantic revision would simply involve memorizing the exceptions to the theory, which is undesirable with respect to generalizing to novel data.

Inductive Logic Programming Representation is Horn clauses Builds rules using background predicates Rules are potentially much more expressive than attribute-value representations

Example Results Rules for family relations from data of primitive or related predicates. uncle(A,B) :­ brother(A,C), parent(C,B). uncle(A,B) :­ husband(A,C), sister(C,D), parent(D,B). Recursive list programs. member(X,[X | Y]). member(X, [Y | Z]) :­ member(X, Z).

ILP Goal is to induce a Horn­clause definition for some target predicate P given definitions of background predicates Q i. Goal is to find a syntactically simple definition D for P such that given background predicate definitions B –For every positive example p i : D  B |= p –For every negative example n i : D  B |/= n Background definitions are either provided –Extensionally: List of ground tuples satisfying the predicate. –Intensionally: Prolog definition of the predicate.

Sequential Covering Algorithm Let P be the set of positive examples Until P is empty do Learn a rule R that covers a large number of positives without covering any negatives. Add R to the list of learned rules. Remove positives covered by R from P

This is just an instance of the greedy algorithm for minimum set covering and does not guarantee that a minimum number of rules is learned but tends to learn a reasonably small rule set. Minimum set covering is an NP­hard problem and the greedy algorithm is a common approximation algorithm. There are several ways to learn a single rule used in various methods.

Strategies for Learning a Single Rule Top­Down (General to Specific): –Start with the most general (empty) rule. –Repeatedly add feature constraints that eliminate negatives while retaining positives. –Stop when only positives are covered. Bottom­Up (Specific to General): –Start with a most specific rule (complete description of a single instance). –Repeatedly eliminate feature constraints in order to cover more positive examples. –Stop when further generalization results in covering negatives.